How do we know they haven't gone with a higher energy density chemistry to lower the cell count and then just put "mass dummies" in the slots where the rest of the 6831 would have been? That might be a more cost effective way to lower overall cost.
They are going to build N of these upgrade packs, where N is <= 2,500. Suppose it takes $10 million to do the engineering - that's $4,000 per car at best, and potentially a lot more than that depending on N.
As for the actual value of N, a lot depends on the cost of the upgrade. If it is $15,000 then I would expect a large number of owners will upgrade. But even at that price many owners will think, "nah it's working fine now, I'll just run it until the battery starts to die." That means the upgrade could be delayed potentially many years. Plus there are a number of these cars sitting in collections not getting many miles.
If the cost is $40,000, then the upgrade numbers could be very weak. I certainly wouldn't bother upgrading my car at that price. My pack is down about 4% from brand new, and for longer trips I usually take the Model S.
Sure, there are many people who will want to get the upgrade as soon as it is released at almost any cost, but there are many more who will wait until they've "used up" their existing pack. Assuming a "reasonable" price for the upgrade, I'd say an
optimistic projection would be 500 to 1000 upgrades the first year, and maybe 200 a year after that.
The above strongly suggests that they will go a route that minimizes the NRE cost. Simply swapping new cells into the existing architecture and tweaking the firmware is the lowest-cost route.